Friday, 29 October 2010

EuroQSAR 2010

I thought this would be a good photo to start the post on EuroQSAR 2010. It really was great fun to be in Rhodes and to catch up with a lot of folk whom I've not seen for a while. The photo was taken at Delphi the day after the conference ended. This the Temple of Apollo where the priestess in residence would inhale hot gases and make predictions... of course nothing like QSAR!

You may well ask what a conference on QSAR has to do with FBDD so I'll try to make the connection clearer. The central problem in QSAR is prediction of affinity so it's a good idea to maintain awareness of the field if you're planning to exploit protein or ligand structures in selection of fragments for screening. Also if you're planning to analyse and design compound libraries or assess druggability then it's useful to know a bit about the molecular descriptors and data analytic methods that form the basis of modern QSAR.

The ultimate goal in QSAR is to start with a molecular structure and predict the physiological effects of the compound. In order to to this you need to be able to predict the extent to which the drug binds to its primary target and a number of anti-targets. You can calculate the extent of binding from the affinity of the drug and its unbound (e.g. to plasma protein) concentration in the vicinity of the target. With typical dosing the unbound drug concentration is a function of both time and location (e.g. intracellular versus extracellular) within the body. If that sounds unbearably complex then I should warn you that it can get a whole lot worse because binding might be slow and you might also need to worry about things like reactive metabolites, isoforms and how toasted the patient got in the pub the night before.

I believe that we're a very long way off seeing this goal achieved. Even when the structure of a protein is known, prediction of affinity for an arbitrary ligand is just not accurate enough. Prediction of unbound concentration at arbitrary location in the body is equally difficult although for some targets it may be sufficient to know the unbound plasma concentration. Nevertheless it is possible to build useful models for some of the pieces in this jigsaw (e.g. binding to plasma protein; IC50 for series of chemically similar receptor antagonists) especially when the process in question is strongly influenced by lipophilicity. QSAR models can be local (i.e. only applicable within a restricted regions of chemical space such as a series of analogs) or global (applicable to any arbitrary molecule). My view is that QSAR models presented as global are frequently ensembles of local models and I have expressed this opinion in print.

I guess that something should be said about the talks after such a long-winded introduction. Some of the speakers have made their lecture slides available. I should first point out that I managed to miss the only talk specifically on FBDD because I was retrieving my camera from my hotel room so that I could photograph a friend doing her poster presentation. The inaugural lecture is given by Hugo Kubinyi and although I saw 'The long road from QSAR to virtual screening' two and a half years ago at an OpenEye meeting in Strasbourg, it is still amusing to see polar surface area and connectivity descriptors cop some flak.

Anthony Nicholls delivers a stimulating lecture entitled 'Information Theory & QSAR'. Nearest neighbor models crop up in more than once in this talk and I liked the suggestion that in validation we should be 'making tests NN resistant'. Nearest neighbours also make an appearance in Han van de Waterbeemd's talk (Assessing Drug Safety and Efficacy through ADME predictions) in the context of 'correction libraries'. The idea behind a correction library is to see if there are systematic errors in the predicted values of a property for near neighbours of the compound for which you're making a prediction. If so the differences between the values measured for the neighbor and predicted for them by the model can be used to correct predictions for new compounds. Of course Orwell would have said that all QSAR models are global...

I enjoyed (not least because he did not appear over-awed by the illustrious inaugural lecturer) the lecture by Alex Tropsha entitled 'Novel Approaches to Chemical Toxicity Prediction Relying on the Entire Structure- in vitro in vivo Data Continuum'. Attempting to sum up the talk in one sentence, I'd say that this was a view of how QSAR modelling might be used to integrate diverse data types that vary in complexity and noise level. One comment that I captured from the 'Notes on chemical descriptors' slide (#14; it's marked 'de-Ku' at bottom left) was that 'descriptors are designed to reflect uniqueness of a molecule in comparison with other molecules'. What does this say about nearest neighbours, I wonder...

Jordi Mestres makes the point in his talk, entitled 'Ligand-based Approaches to In-Silico Pharmacology', that we should STOP using the word 'polypharmacology'. I couldn't agree more although I think the term 'pharmacodynamics' is a couple of orders of magnitude more meaningless. This lecture is about predicting affinity across a range of GPCRs and how these affinity profiles might be used, for example, to anticipate side effects of drugs. Intracellular targets will add complexity because less will be known about unbound concentration of drug in the vicinity of the target.

There's also a session on design of agrochemicals kicked off by Klaus-Juergen Schleifer. All molecular design is subject to constraints and it is always educational to see how people designing molecules for different purposes deal with the constraints that apply to them. Designers of agrochemicals need to deal with different species of plants, fungi and insects in a chemical-unfriendly (e.g. bright sunlight, rain) environment. Pharmaceutical QSAR modellers may find that they learn more in an agrochemical session than a pharmaceutical one.

I must confess to being less than alert on the Friday morning and this may have something to do with over-indulging at the conference dinner and ending up on the beach at 2AM (it seemed a good idea at the time but then it always does). Eric Martin describes how QSAR methodology can be used to integrate affinity and protein structural data for protein kinase inhibitors. Tudor Oprea's lecture (Computer-Aided Drug Re-purposing) is notable because of the use of experimental pharmacokinetic data (this is typically available for marketed drugs) to get a handle on unbound plasma concentration. Aspiring systems biologists, take note.

I've taken a look at some of the talks and it's a good time to summarise before getting on to the fun bit where I share some pictures. Firstly I do not see prediction of affinity for arbitrary molecules (e.g. when there is no measured data for analogs) something we can currently do in a useful and general manner. Secondly, I don't see model validation as a solved problem and a session on the subject is something that the Scientific Committee may wish to think about for EuroQSAR2012 in Vienna. Thirdly remember that, "A theory has only the alternative of being right or wrong. A model has a third possibility: it may be right, but irrelevant" (Manfred Eigen).

The excursion to Lindos provides an excellent opportunity get some pictures. First to be papped is fellow AZ-escapee Han.

I sneak up on Anna (who also doesn't work at AZ any more) as we wait for the bus.

Dick is the man behind CoMFA and Topomers and I've lost count of the number of conferences which we have both attended.

Unfortunately I can't get Yvonne to look at the camera. To her right is Cynthia who co-chaired the session on descriptors in which Yvonne and I both did talks. Andreas and Ylva are smiling because I've told them that the chef has got rotted herring on the menu for the conference dinner.

I finally catch Yvonne as we're waiting for the bus back. I'd not seen Eric since the mid-90s so it's good to get the two of them together in this photo.

David (make sure you're in a country with an almost worthless unit of currency should you presume in his presence otherwise prepare to pay out big time) and Frank have been in this business a long time.

Portraits from the conference dinner.

David is animated but Ant looks in need of a dose of the Poisson-Boltzmanns.

Dimitris appears strangely unconcerned to be in the company of two Transylvanians who are discussing anticoagulant QSVRs (Quantitative Structure Viscosity Relationships) as they admire the tone of his carotid arteries.

Graduate students. Birte (4th from left) did a talk and both Juliana (2nd from left) and Andrea (3rd from left) had their posters selected for oral presentation.

Party animals.

Friday, 1 October 2010

Molecular Interactions

Molecular interactions are an important part of the theoretical framework of modern drug discovery and studying them is a great way to increase your understanding of physicochemical principles of molecular design. One view of molecular design is as a process of tuning the interactions of molecules with the different environments in which they exist. Needless to say, knowledge of molecular interactions is particularly valuable in FBDD. Three Roche scientists have recently published “A Molecular Chemist’s Guide to Molecular Interactions” which should be of interest to anyone working in molecular design and other bloggers ( Derek | Joerg ) have already highlighted the article.

The authors cover plenty of ground and everybody should find their favourite molecular interactions discussed. Given the recent LinkedIn discussion on the value of measuring enthalpy and entropy changes associated with binding, I was pleased to see that the authors noted that interpretation of these quantities is typically difficult. The discussion of cooperativity was useful because we often assume that contributions of interactions to binding are additive.

One of my favorite interactions is halogen bonding and I am pleased to see it discussed in some detail. The halogens all confer a degree of hydrophobicity on a molecule and the heavier halogens (i.e. not fluorine) also exhibit an ability to interact with hydrogen bond acceptors that increases with atomic number. Although this class of interaction is sometimes thought to have been discovered in recent times, it’s actually been around a long time. I can remember learning, as a schoolboy in Port of Spain in the mid-1970s, about why iodine is more soluble in aqueous potassium iodide than in water. I developed this theme a bit more in a light-hearted survey of halogens at EuroCUP in 2008 which starred both Bismarck and a medical writer by the name of Bouchardat who was active in Paris when Pincess Victoria became Queen Victoria. I'm not sure if they ever caught the notorious Parisian dog-poisoner.

Something that I found disappointing was that there was not a lot of information about how much additional affinity you’re likely to get by making the different interactions. This should not be seen as a criticism of the authors who have carried out an impressive trawl of the literature. It’s just disappointing that the information is not available in the current literature base.

I’ve some comments to make on the discussion of hydrogen bonds. There is a widely accepted view that the maximum contribution to affinity that a hydrogen bond between a neutral donor and acceptor can make is just over a log unit and here’s what the Roche authors had to say on the subject:

“Hydrogen bonds always convey specificity to a recognition process but do not always add much binding free energy. Desolvation of the donor and the acceptor must occur for the hydrogen bond to form, such that the effects of hydration and hydrogen bond formation nearly cancel out”

I certainly agree that in some cases the contribution of hydrogen bonds to affinity will be minimal. However, the dataset from which that figure of just over a log unit was derived is actually quite small and not especially diverse in terms of donor-acceptor pairings. I believe that if you can form the hydrogen bond deep in a binding pocket then it can make more than the widely accepted maximum contribution to affinity. We recently published inhibition data which included the example of aza-substitution of a pyridine ring resulting in increases of potency of about two log units against Cathepsins S and L2. Some caution is required in interpreting these results because we didn’t have the relevant crystal structures and the inhibitors are racemic. However, I do believe that these results should make us question the prevailing view of the maximum contribution that a hydrogen bond between a neutral donor and acceptor can make to affinity.

I’ll say some things about the discussion of the hydrogen bonding of sulfonyl groups because it should get you thinking a bit. The authors state that:

“Only 30% of the sulfones and sulfonamides form hydrogen bonds. This raises the question of which type of interaction this functional group prefers.”

I’m not sure that I agree with the second sentence and would be interested to know how many of these sulfones and sulphonamides actually had the opportunity to accept a hydrogen bond. If no hydrogen bond donors are present in a molecule then you can’t really blame the sulfonyl oxygens for making contact with aliphatic carbon in the solid state since that's going to be a better option than getting in the way of the sulfonyl oxygens of a lattice neighbour. Even when a donor is present in the molecule, the favoured interaction may well be with a stronger acceptor than the sulfonyl oxygens.

The authors also took a look at the environments of sulfonyl groups in the PDB and here’s what they had to say.

“Notably, of the sulfonyl groups situated in a hydrophobic environment in the PDB, only 36% are found to interact simultaneously as a hydrogen bond acceptor but 79% of the hydrogen-bonded sulfonyl groups are found to interact simultaneously with a hydrophobic group. These findings clearly indicate a dual character of the weakly polar sulfonyl groups as a hydrogen bond acceptor and as a hydrophobic group.”

I simply don’t buy this idea of sulfonyl oxygens having a dual acceptor-hydrophobic character. Hydrophobicity is a statement of aquous solvation characteristics. It will be easier to place a weak acceptor in a hydrophobic environment than it will to place a strong acceptor there. However, if it’s an acceptor, it’ll still prefer the aqueous environment. Think about the consequences of one of these oxygens accepting a hydrogen bond. When an oxygen atom is hydrogen bonded its ability to accept a second hydrogen bond is likely to be reduced because the donor polarises the acceptor oxygen. Also the second donor will also experience repulsive secondary electrostatic interactions with the existing donor. Provided that the oxygen can still accept a hydrogen bond, it will not be too ‘distressed’ (apologies for anthropomorphising) to be in contact with hydrophobic surface. If you’re interested in this sort of thing then take a look at our article on alkane/water partition coefficients to see how accepting a hydrogen bond (from methanol which we used to model octanol) is likely to affect the ability of carbonyl oxygen to accept a second hydrogen bond.

To be fair, the authors do recognise that accepting a hydrogen bond might affect the probability of a sulfonyl oxygen atom making contact with hydrophobic surface. However, this hydrogen bond donor doesn’t need to come from the protein or a water molecule that the crystallographers can see. How many of the sulfonyl oxygen atoms which lack ‘visible’ hydrogen bonds are sufficiently solvent-exposed to accept hydrogen bonds from ‘invisible’ solvent water molecules? I’ll leave it to you the reader to think about whether analysis of the CSD (as opposed the PDB) has any relevance to hydrophobic interactions.

I’m now going to wrap up with what will be seen by some to be nitpicking although that is not my intention. This is what the authors have to say about QM calculations and hydrogen bonding:

“Where experimental data are not available, acceptor strengths can be obtained from quantum chemical calculations.”

My first criticism of this statement (which is getting very close to nitpicking although as The Blogger I’m allowed to do that) is that quantum chemical calculations can be used to predict donor strengths as well. Like they might say in Buenos Aires, it takes two to tango. My second criticism is that rather than talking about generic “quantum chemical calculations” the authors could also have mentioned that electrostatic potential is a useful predictor of both acceptor and donor strength. I have to declare an interest here as author of reference 98d but I do believe that the effectiveness of electrostatic potential as a predictor of donor and acceptor strength is more important than whether it was calculated quantum mechanically or classically. It tells us something about the nature of the hydrogen bond.

That brings us to the end of my review. The article is definitely a good read and a valuable contribution to the field. To put it bluntly, you need to know this stuff if you want to succeed in FBDD. I’ve flagged up the issue of sulfonyl oxygen hydrogen bonding to hopefully make you think a bit and maybe even generate some discussion. Feel free to make comments of your own.

Literature cited

Bissantz, Kuhn & Stahl, A Medicinal Chemist’s Guide to Molecular Interactions. J. Med. Chem. 2010, 53, 5061-5084 DOI

Davis & Teague, Hydrogen Bonding, Hydrophobic Interactions, and Failure of the Rigid Receptor Hypothesis Angew. Chem. 1999, 38 736-749 DOI

Bethel et al, Design of selective Cathepsin inhibitors. Bioorg. Med. Chem. Lett. 2009, 19, 4622-4625 DOI

Toulmin, Wood & Kenny, Toward Prediction of Alkane/Water Partition Coefficients. J. Med. Chem. 2008 51, 3720-3730 DOI

Kenny, Hydrogen bonding, electrostatic potential and molecular design. J. Chem. Inf. Model. 2009, 49, 1234-1244 DOI